[netCDF #XDD-271000]: nf90_put_att with LOGICAL attribute

David,

If by "booleans seem to be supported" you mean that the netCDF data
model supports attributes with no values (and thus no type), that's
almost right, except attributes still have to be given a type, and
boolean is not a supported type.  If the mere existence of an
attribute is all that's important, you can assign it a numeric value
such as 1, or a string value, such as "" (the empty string).  The
latter is the convention supported by ncdump for attributes that are
created with no values (which the APIs currently allow).

If you want to store a mask array efficiently, there is a convention
for doing that that has been accepted into the CF conventions, and
will appear in the next version of the CF conventions document.  It
uses just one bit for each flag value, and permits both exclusive
flags as well as enumerated flags:

  http://cf-pcmdi.llnl.gov/trac/ticket/26

There is currently no API in netCDF to support that convention, so if
you want to look at that and develop something suitable, it might be
useful.

One thing to remember about new data types in netCDF is that it should
be possible to implement support for them portably in all the language
interfaces for netCDF (C, Fortran-77, Fortran-90, C++, Java, Perl,
Python, Ruby, etc.) so that data written with any language interface
can be read and interpreted correctly with any other language
interface.  This principal is violated currently by NC_UINT64, which
is supposed to represent unsigned 8-byte integers, not available in
all the above languages, but I'd like to not see any more such types
that can cause problems for interoperability.

Adding a new boolean primitive type to the netCDF data model might be
a good idea for a future release, but seems quite ambitious to add as
patch to the current release, if it covered more than just the
Fortran-90 interface.  We do consider patches, however.  What might be
more useful at this point is an independent evaluation of the adequacy
of the proposed masking convention or an implementation of auxiliary
APIs to Unidata's libcf to implement it, or even to check for
conformance with the rules proposed for this convention.

Anyway, thanks for the offer of help!  I agree that boolean masks are
useful, and we want it to be easy to represent data that uses boolean
masks in netCDF.

--Russ

Russ Rew                                         UCAR Unidata Program
address@hidden                     http://www.unidata.ucar.edu



Ticket Details
===================
Ticket ID: XDD-271000
Department: Support netCDF
Priority: Normal
Status: Closed